12mo ago

Video game actors speak out after union announces AI voice deal

www.videogameschronicle.com Video game actors speak out after union announces AI voice deal | VGC

SAG AFTRA claims 80% of its members approved the…

According to SAG AFTRA, the deal will “enable Replica to engage SAG-AFTRA members under a fair, ethical agreement to safely create and license a digital replica of their voice. Licensed voices can be used in video game development and other interactive media projects from pre-production to final release.”
The deal reportedly includes minimum terms and the requirement for performers’ consent to use their voice for AI.
However, several prominent video game voice actors were quick to respond on X, specifically to a portion of the statement which claims the deal was approved by “affected members of the union’s voiceover performer community.”
Apex Legends voice actor Erika Ishii wrote: “Approved by… WHO exactly?? Was any one of the ‘affected members’ who signed off on this a working voice actor?”

72 comments

This is the one career I wanted to do something in. I started voice acting years ago.
Honestly just seems so pointless lately. Half the time voice over artists are not even recognized or even paid a decent amount. Now they wanna just copy and paste our voice.
I'm tired.
- Even the recognized ones don't get paid well. Weird that SAG would sign off on this so easily when SAG actors and writers striked for months with AI being a big part of the issue.
  
  Because voice actors are not considered real actors by SAG-AFTRA. Despite the fact that doing voice over work is typically far more grueling than being in front of a camera. I've never met a VOA who wasn't looked down on by other SAG members. Even by fucking extras. "You just stand in a booth and read lines."
- I've done voice acting before, and honestly, you're right on the money.
  So many people don't even notice if the voice work in a piece of media is good/bad.
  Playing one game, several of the main characters sound like they were recorded in completely different rooms.
  And I'm sat here like a mug with a deadened setup, wondering why I bothered.
  When that little detail is paid, I can see games absolutely jumping on the machine generated bandwagon.
  
  I've been trying to figure out a deadened set up for a while. My closet isn't big enough as a gay man, no. I'm not going back in. I need one of those lil curtain thingies that create a dead space.
  But yeah I agree. Resident Evil 4 killed me for that. So many of the characters sounded amazing and then Ada just stuck out like a sore thumb. I don't mean because of a poor performance or anything either. I wasn't totally happy with it but I blame the game director for that, not her for doing what they asked of her. But everyone else was done in a professional set up and her quality is NOTICABLE different. It fucked up the entire DLC for me and I haven't been able to really enjoy it.
  Doing voice acting and then hearing all the ways people mess up in stuff is astounding. Not that I don't, mind you. I'm not perfect. But my god do you notice all the major errors in every game out there once you start working a bit in the field.
- Wish there were a way for you to post examples without doxxing yourself. Ah well!
  
  I occasionally record readings and narrations of dumb things on Mastodon but they're usually just super quick. Like so quick that I don't even bother fucking with noise reduction or anything. Just get bored and go "I LIEK THIS POEM"

I really hope this doesnt take off. I tried out Star Trek: Infinite and the tutorial uses an AI voice. It just sounded bizarre and jarring, completely took me out of the experience.
- Cant wait to have the voices removed after 3 years due to copyright issues
- This deal solves the problem you're encountering, because it allows game companies to use real voices to generate dialogue. It will sound a hell of a lot better than the 100% AI generated voices you dislike.
  And it will protect voice actors' jobs because the deal effectively requires new contracts for each use out of scope of the previous contract (i.e., the "opt out" language), and it encourages game companies to continue to rely on voice actors rather than switch to 100% AI generated.
  Without this deal, game devs will just go 100% AI (and the tech will improve dramatically), and within a year or two, game voice actors will have no jobs to contract.
  This is especially important in light of the trend toward dynamically generated dialogue in RPGs, etc. Without allowing an AI to train on real voice actors, dynamically generated dialogue will have to be 100% AI generated (no human voice involvement).
  Voice acting in all fields is already a diminishing market because of AI generated voices. One of my coworkers had to get a job where I work because his VA jobs basically dried up. This agreement stanches the bleeding by permitting the use of AI trained on VAs (but only allowing use on a per-contract basis). Without that permission, AI would just be trained on open source / freely available voice samples, and there would be no contracts, and VAs would just .... not exist anymore.
  
  I disagree with it "solving" the problem. I've yet to hear an AI voice that actually works/sounds like an actual person. I've heard sentences or two that are somewhat passable at times, but never enough for actual dialogue. Regardless, your entire comment also does not address the issue presented at all, that voice actors did not agree with this deal.
  
  within a year or two AI actors will have no jobs
  extreeeeeme doubt. The moment an AI has to inflect emotion it really fucks it up. You'll spend 5 hours and $200 of compute costs getting it to say "Great, thanks" sarcastically, when an actor could do it in a single take as part of doing the entire script.
  
  I agree. The key factor is getting this settled before some smart people get this working seemlessly. It's stupid to hear that there wasn't any unionised info decisions for a union though. I guess you ask the union to speak for you but it's the unions job to speak back.
  
  Blah blah blah
  And this deal was vetted and approved by which working voice actors?
  
  Interesting, thanks for the clarification. Have to admit, that does sound better, at least in terms of quality
- Speaking of Star Trek and AI voices... Majel Barrett supposedly recorded her voice so that it could be used in the future by software to make her talk again.
  So fuck Google Assistant or whatever. Where's my Enterprise Computer app for me to talk to?
  
  Can't even configure assistants to call them "Computer"
- The finals uses ai voices for announcers and I could not tell the difference so there's definitely something there. I think it works in that setting because the inflections is so set.
  But as a emotional character maybe not so much

This is one sector where I am actually happy for AI to be available. I want to play a game where the NPC's can say my character name.
That being said, I also want the voice actors to be compensated fairly. Maybe the guilds can set up a deal where using someone's voice for training data is included.
- I feel like the solution is pretty simple: if you want to AI copy someone's voice and put it in your project, you have to hire them and pay them as normal, and they have to give consent to let the AI use their likeness.
  Otherwise it's theft.
  
  And this has to be on a per-game basis, to. Studios licensing a voice in perpetuity will eventually come back to the same issues.
  For AI to truly be a net benefit to our society, it should be used as a tool by the artists to augment the output from the artists. It shouldn't be a way of replacing them.
  If a voice actors job goes from recording each and every line to recording samples for AI and helping to tweak the output, that's fine. But the compensation stays the same.
  That's how it improves our world. Makes the human's job easier without replacing them or affecting their compensation.
  The way it's currently on track to be used is how it improves the lives of the wealthiest at the expense of everyone else. No amount of futurist techno-jerking should distract from that. These are not tools for us to benefit from in any significant sense.
  
  That's... what this agreement proposes.
- They don't need AI to say your character's name, text to speech works for that.
  
  Yeah, but matching TTS to a character’s voice is where you would need it.
  
  Sure, it's just that this specific text to speech voice is created by an AI via training data via voice samples.
  AI is more than just ChatGPT, it's an algorithm that can be applied to a lot of different things.

I think it would be better for all involved if we figure this out now. Existing Voice Actors should not have their performances used without their explicit consent. Any performances used by current or past voice actors must have explicit consent and compensation. "New" voices generates by AI must be sufficiently differing from existing performances and any existing performances used in the generation must have consent by the original voice actor.
- Existing performances must not be used to train models. If you wish to train a model you should need explicit consent and hire an actor to record such data. The actor should also receive royalties when the resulting model is used for a commercial purpose.
  See, minus the royalty part (in most cases) this has been how VOCALOID, SynthV and the like has more or less operated for two decades now.
- New creations from existing training data from an actor should have some type of royalties involved. The complication with that is the AI tools are largely a black box and it can be murky on where things come from.
  
  Thats a solid and necessary addition. This comment section alone has already done more than any regulators. Its like they're afraid to at least lay down protective ground rules so VAs can continue to eat. Too much profit salivation.
- New voices generated by AI can't be covered under copyright law, so I doubt they will see much use from corporations.
  
  Yet. Once upon a time we couldn’t patent an organism and yet now GMOs and companies like Monsanto abuse the legal framework.
  
  That's correct, but it's important to distinguish something explicitly here. The voices may not be copyrightable, but the dialogue is, as long as it's not also generated by AI (i.e., dynamically generated). Also, the trained model that generates the voice is still proprietary: only its product (and only the sound itself, not the words if the speech is from a script) can be openly used.

I feel like this is really a consequence of what many called the "bad deal" the SAG/AFTRA merger was years ago. When the union can effectively exclude you from the bargaining process and arbitrate you to it, what's the point? They're behaving like a cartel, and not like a union. This is not praxis, brothers and sisters!

This solution shouldn't be that hard, just create an AI model for every individual "voice" or character and then license it for use or receive royalties on it.
They'll probably use it as filler for side dialogue and then have the VA do all the main lines to really nail the human presence, since AI isn't as good at emotional inflection.
- Honestly this would be a good method. Limit AI voice acting to only single use NPC such as Town folk when you visit a town and then have like shopkeepers or party members or the main character actual voice. You aren't expecting much out of those temporary characters anyway so them having weird Oddity voices isn't going to be super jarring for the environment. Plus it will help you as the player realize which characters are supposed to be part of the story and which ones are there for just Scenic effect
  
  I mean even main characters could have AI generated dialogue, you have the VA do the voice until there is enough sampling data to train a model on, and then you can use that for any small or side content.
  Then just have that characters AI model be owned by the actor and use of the voice gives them royalties for it. Then you can supplement actual lines with generated banter, etc. While still giving the VA compensation for their voice and likeness.

I'm torn, because on the one hand, the logistics of constantly recording new lines for minor stuff is really annoying. Imagine you're playing a live-service game that really needs a certain balance patch, but that balance patch is reliant on a very slight change to a voice line (for instance, reducing the time it takes for a character to perform a special attack. To take an Overwatch example, maybe a certain archer is voicing his ultimate ability too quietly). Having to call someone in just for that is costly and unproductive.
But, we're talking about delivering the source of someone's work and livelihood (as well as all their creative influence, exaggerative tones, and delivery) into an algorithm. The line where it would go beyond convenience into worker-reduction efforts is going to be hard to draw.
I would rather that the voice actor retains the rights to their voice, even if it's put into an AI algorithm. Thus, if the developers want to make a small change to a voice line, they still need to get approval for some AI-generated correction - and the actor would have the right to say "No, that one sounds terrible. I'm only going to agree to re-delivering this one myself." Similarly, actors could approve limited sets of explicitly-defined live AI usage, for instance pronouncing the player's name. Granted, some companies would become annoyed at actors being too inflexible, just like they have disagreements with actors today.
I'm definitely worried about too much signing-over of voice identity. I think it's very easy to cut humans out of the equation that way, which not only damages the health of the industry, but also reduces creative output.
- Counter point. That live service game makes a billion dollars a year and can afford to spend the time and money to re-record that line.
  If we're gonna use Ai, it shouldn't be to make massive corporations' lives easier and more profitable at the expense of workers and quality.
  
  In my mind, they should be paying the actor the same for the new lines regardless of whether they opt for them to come back in and re-record or use AI to generate the new line. The actor's product (their voice) isn't worth any less, but the company could save money by streamlining the creation of a new line through simplified logistics. This way the company has some benefit while preserving the actor's livelihood.
  Of course there's no way these companies are going to want to pay full price for these new lines, since it's an obvious point where they can pressure performers to accept a lower rate.
  
  While I agree, the corpos dont and will fight tooth and nail to cut the cost anyway.
  So unless the US gets the stones to collar and muzzle these businesses (they wont) we have to work around these monsters who will bite your arm off to skip lunch

I’m hoping this also means that voice actors can choose not to enter such a deal?
- It does, yes. And they can also choose to opt out of future uses of their voice in the AI trained model. Which essentially means that their contracts are on a per-project basis, rather than allowing the game developer to force them to contract for the current project and any future use of the model by that game dev.
  
  The way I see it, if they want to train models on someone’s voice they should hire them specifically for that purpose. Ergo, clips that are used in production should not be used for training voice models.
- Yes but it also means that the studios can only hire someone who will sign off on it.
  
  Not if they want a particular name, like Elias Toufexis seems pretty against having his voice cloned.

To be fair, in November SAG-AFTRA did also make a deal for movies and TV for AI likenesses being used in projects. However, I do think that video games and voice acting in general are a bit of a different beast since they're often already overlooked and underpaid.
I mean the union fought tooth and nail and had a vote on the AI deal for their silver and big screen members and while I can kind of understand that the union leadership probably went, "Well we have a blueprint from our last vote, we should be good to just use it again" type of thing, I can understand why the VA members would be a bit upset and feel disrespected in not having the same courtesy applied with allowing them to have the same review and vote.

I sympathize with the voice actors but at the same time I think this is a losing battle. I expect AI voices to be widespread and employment opportunities for voice actors to diminish (although I think high-budget games will still use human voice actors for a while). Maybe being open to AI is actually the best case scenario for getting at least some of the money involved.
- So tech outpaces legislation, as it is wont to do since legislation is notoriously slow, and so because of that our reaction should be to throw our hands up and not even try? Perhaps you don't sympathize as much as you think you do.
  
  I don't think this is a case of tech outpacing legislation because I expect that ultimately legislation will be rather favorable to tech. There's too much money to be made using AI for the government to extend copyright protection to training data.
  (Plus, I sympathize with voice actors in the sense that I'm sad that a lot of them will lose their jobs, and that's independent of what I think about AI development and copyright law.)
  
  They're not giving up though, what they're doing is getting ahead of it. Assuming their deal is favorable for their members, they're making it so that anyone who wants SAG-AFTRA synth voices has to go through their contracted company which they have collective bargaining power or strike an equal or better deal. Along with blacklisting companies from SAG-AFTRA work that use non-union synth voices.
  This is way better than leaving actors on their own to bargain with companies, which would have definitely happened. Rather than have companies wear individuals down and drive pay down, they get to dictate the terms, together.

72 comments